1. Please paste your SMILES list, Drug Name list or upload a file here. (example) Each query can handle 50 compounds at most.
ADMET, i.e. Absorption, Distribution, Metabolism, Excretion and Toxicity, is one of the important domains in drug research & development pipelines. Drug candidates have to be systematically optimized by balancing multiple key ADMET properties. Computational predicting ADMET properties makes it possible to assess ADMET properties virtually and early in drug pipelines, helping to speed up the optimization process by better experimental design as well as saving resources and time.
By combining different drug pipelines’ knowledge, over 40 important and common ADMET properties (Figure 1) were collected, and deep learning predictive models were then constructed for GHDDI’s Free ADMET Prediction Service. For each ADMET property, positive, toxic, or active compounds were labeled 1, while negative, nontoxic, or inactive ones were labeled 0. Different training sets were prepared using different cutoffs, e.g. Papp>=10*10^(-6)cm/s for Caco2, CLint>=20mL/min/kg or T1/2<=30min for HLM. Based on more than 550,000 historical experimental records, about 50 predictive models were constructed and validated in 5-fold cross validation. Performance was measured using AUC and AuPR (Figure 2). (More details about AUC and AuPR). Both best AuPR models and best AUC models were online for different demands.
This service helps to systematically profile and assess your own compounds’ ADMET properties automatically. The report ADMET Druglike Score radar chart is generated by averaging different properties’ predictive scores after category standardization. For each end of radar graph, the higher the score, the better or more drug-like the compound is considering ADMET properties.
Each upload or submission can handle 100 compounds at most. All the models in this service will be maintained and updated regularly.